Mesoscopic Energy Consumption Prediction for Electrified Traffic Flows Using a Multichannel CNN
Electrifying road traffic creates a need for segment-level energy estimates that can support charging-demand assessment and infrastructure planning. This study constructs physics-model-derived reference energy labels by applying an electric-vehicle longitudinal-dynamics model to Next Generation Simulation trajectories under a hypothetical fully electrified light-duty fleet scenario. A multichannel residual convolutional neural network (CNN) (CAMMCRN) is then developed to estimate the reference energy from mesoscopically observable spatiotemporal traffic features. Experiments on high-occupancy vehicle and general-purpose lanes show that, at the 15 m × 15 s resolution, CAMMCRN achieves mean absolute percentage error values of 12.77% and 9.68%, corresponding to relative reductions of 56.0% and 62.7% compared with the multilayer perceptron benchmark. The framework supports traffic-energy estimation when detailed per-vehicle parameters are unavailable; however, its outputs remain conditional on the representative electric vehicle parameters used to generate the reference labels.
Authors
- Jiawen Li (ORCID: https://orcid.org/0009-0008-2549-2845)
- Pengfei Li
- Jianhong Wu
Institutions
- Xi’an University of Posts and Telecommunications (CN)
Publication Details
- Journal
- Journal of Transportation Engineering Part A Systems
- Published
- 2026-11-18
- DOI
- https://doi.org/10.1061/jtepbs.teeng-9759
- Primary Topic
- Electric Vehicles and Infrastructure
- Type
- article
- Field-Weighted Citation Impact
- 0.00
Funders
- Education Department of Shaanxi Province
- National Office for Philosophy and Social Sciences
- Key Science and Technology Program of Shaanxi Province